Download 2D SMARTCyp Reactivity-Based Site of Metabolism Prediction for

Article
pubs.acs.org/jcim
2D SMARTCyp Reactivity-Based Site of Metabolism Prediction for
Major Drug-Metabolizing Cytochrome P450 Enzymes
Ruifeng Liu,*,† Jin Liu,† Greg Tawa,† and Anders Wallqvist†
†
DoD Biotechnology High Performance Computing Software Applications Institute, Telemedicine and Advanced Technology
Research Center, U.S. Army Medical Research and Material Command, Fort Detrick, Maryland 21702, United States
S Supporting Information
*
ABSTRACT: Cytochrome P450 (CYP) 3A4, 2D6, 2C9, 2C19, and 1A2 are the most
important drug-metabolizing enzymes in the human liver. Knowledge of which parts of a drug
molecule are subject to metabolic reactions catalyzed by these enzymes is crucial for rational
drug design to mitigate ADME/toxicity issues. SMARTCyp, a recently developed 2D ligand
structure-based method, is able to predict site-specific metabolic reactivity of CYP3A4 and
CYP2D6 substrates with an accuracy that rivals the best and more computationally demanding
3D structure-based methods. In this article, the SMARTCyp approach was extended to predict
the metabolic hotspots for CYP2C9, CYP2C19, and CYP1A2 substrates. This was
accomplished by taking into account the impact of a key substrate-receptor recognition
feature of each enzyme as a correction term to the SMARTCyp reactivity. The corrected
reactivity was then used to rank order the likely sites of CYP-mediated metabolic reactions.
For 60 CYP1A2 substrates, the observed major sites of CYP1A2 catalyzed metabolic reactions
were among the top-ranked 1, 2, and 3 positions in 67%, 80%, and 83% of the cases,
respectively. The results were similar to those obtained by MetaSite and the reactivity + docking approach. For 70 CYP2C9
substrates, the observed sites of CYP2C9 metabolism were among the top-ranked 1, 2, and 3 positions in 66%, 86%, and 87% of
the cases, respectively. These results were better than the corresponding results of StarDrop version 5.0, which were 61%, 73%,
and 77%, respectively. For 36 compounds metabolized by CYP2C19, the observed sites of metabolism were found to be among
the top-ranked 1, 2, and 3 sites in 78%, 89%, and 94% of the cases, respectively. The computational procedure was implemented
as an extension to the program SMARTCyp 2.0. With the extension, the program can now predict the site of metabolism for all
five major drug-metabolizing enzymes with an accuracy similar to or better than that achieved by the best 3D structure-based
methods. Both the Java source code and the binary executable of the program are freely available to interested users.
1. INTRODUCTION
The cytochrome P450 superfamily of enzymes (abbreviated as
CYP) comprises a large and diverse group of proteins with the
heme cofactor.1 In humans, they transform lipophilic drugs to
more polar compounds that can be excreted by the kidneys
and, therefore, play important roles in defining a drug’s
pharmacokinetic profile.2 They also contribute to drug−drug
interactions and metabolism-dependent toxicity issues.3 Among
the CYP enzymes, CYP3A4, 2D6, 2C9, 2C19, and 1A2 are the
most important members in human liver, the principal organ
for phase 1 metabolism and clearance of drugs and other
chemicals.4 Together, the five enzymes metabolize approximately 90% of marketed drugs.5−7 The ability to predict
substrate specific sites of metabolism (SOM) of these enzymes
is essential for rational drug design to mitigate ADME and
toxicity issues of a drug candidate.
The mechanism of CYP-mediated metabolism is complex
and consists of multiple-steps.8 However, there is experimental
evidence indicating that the rate-determining step at least
partially involves hydrogen or electron abstraction from the
substrate followed by oxygen rebound or a concerted
oxygenation via formation of a sigma complex between the
substrate and the FeO3+ complex.8 Chemical reactivity,
© 2012 American Chemical Society
therefore, is an important determinant for the SOM of a
substrate. On the other hand, due to the different shapes and
sizes of substrate binding cavities of the CYP enzymes and their
characteristic substrate recognition features, substrate exposure
to the catalytic moiety may be restricted. As a result, the most
reactive site of a substrate may not be the observed SOM. This
underscores the necessity of considering substrate-receptor
recognition in predicting SOM.
In principle, a promising method for predicting substrate
accessibility to the CYP catalytic moiety is docking simulation.
Many docking studies aimed at predicting substrate sites of
metabolism have been published in recent years.9−12 Major
challenges for this approach include difficulties associated with
proper representation of a protein and proper accounting of its
flexibility, the lack of a score function that does not require
tuning by a user with a priori knowledge for ranking the docked
poses, and insufficient force field parameters to describe
interactions between the substrate and FeO3+ complex.13
Recently, Moors et al. demonstrated an approach to account
for protein flexibility in a CYP2D6 docking study.14 They
Received: March 20, 2012
Published: May 25, 2012
1698
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
likely sites of CYP3A4 catalyzed reactions. For 394 CYP3A4
substrates, the observed SOM were among the top-ranked one,
two, and three sites 65%, 76%, and 81% of the time,
respectively. They are at least the same or better than the
performance of StarDrop21 and MetaSite.22 To predict
CYP2D6 catalyzed SOM, SMARTCyp introduced two 2D
descriptors to correct the energy of each atom, an accessibility
descriptor and a descriptor to account for the impact of the
characteristic receptor − ligand recognition between CYP2D6
and a positively charged ligand.23 When the approach was
applied to predict the SOM of a large number of CYP2D6
substrates, its performance was shown to be better than
available commercial packages.23
Encouraged by the success of SMARTCyp, we wondered if
the same approach could be applied to predict the SOM of
other major drug-metabolizing enzymes, namely CYP1A2,
CYP2C9, and CYP2C19. In this article, we demonstrate that
taking into account the effect of a single key receptor-substrate
recognition feature for each of the enzymes, the SMARTCyp
approach can be extended to provide accurate predictions of
the SOM catalyzed by these enzymes.
generated an ensemble of 1,000 CYP2D6 conformations
starting with the X-ray structure of apo-CYP2D6. Docking
simulations were performed on every conformer of the
ensemble for many CYP2D6 substrates. When information
from the docking simulations was combined with the estimated
site reactivity of the substrates, reliable SOM predictions were
achieved. However, even with the availability of today’s
massively parallel computers, docking simulations using an
ensemble of thousands of protein conformations and
processing a huge number of docked poses are still timeconsuming and not practical for virtual screening of a large
number of compounds.
Moors and co-workers also showed that predictions based on
docking to any single protein conformation are significantly less
reliable as judged by the scores of receiver operating
characteristic (ROC) curves of their predictions.14 This is in
agreement with results of a recent study by Afzelius et al. who
evaluated SOM predictions for CYP3A4 and CYP2C9 based on
docking using Dock and Glide software.15 For some structurally
diverse CYP2C9 and CYP3A4 substrates, they found that the
top-ranked sites, based on docking to a single protein
conformation, were among the observed SOM only 30% to
40% of the time. This accuracy is lower than predictions given
by an experienced biotransformation scientist, which were
approximately 50%.15 Furthermore, it is significantly lower than
predictions based on C−H bond orders of the substrates,
calculated by density functional theory (B3LYP) with a small 321G basis set, which achieved around 60% accuracy.15
Chemical reactivity is the basis of SOM predictions of
CYP3A4-mediated metabolism by Singh et al. who estimated
hydrogen abstraction energies by AM1 molecular orbital
calculations and predicted SOM by the energies and surface
area of the hydrogen atoms.16 AM1 estimation of substrate
reactivity is also the basis of SOM prediction by StarDrop, a
commercial software package, which makes on-the-fly AM1
calculations.17 In addition, StarDrop makes corrections to the
AM1 energies to account for steric accessibility and orientation
effects via models trained using a large number of substrates of
CYP3A4, 2D6, and 2C9, the top three major drug-metabolizing
P450 enzymes.
Another popular commercial package for SOM prediction is
MetaSite.18 It identifies likely sites of metabolism by fitness of
3D structures of a substrate oriented within the catalytic sites of
the CYP enzymes represented by GRID molecular interaction
fields.19 For a large number of CYP3A4 substrates, Zhou et al.
found that the observed SOM were among the three topranked sites by MetaSite in 78% of the cases studied.20
However, the success rate for the observed SOM to be ranked
the highest is significantly lower.15
Generally speaking, 3D molecular structure-based prediction
methods, such as docking into a large number of protein
conformations and/or quantum mechanical calculation of
substrate reactivity at appropriate levels of theory, are
computationally demanding and not easily applicable to virtual
screening of a large number of compounds. An alternative is to
estimate the reactivity of molecular fragments by high-level
quantum mechanical calculations on representative molecules
and assign reactivity to different sites of a substrate by matching
structural patterns. This is the approach used by SMARTCyp, a
2D substrate structure-based SOM prediction method.21 To
predict SOM of CYP3A4-mediated reactions, the SMARTCyp
energy of each potential site is corrected by an accessibility
descriptor. The corrected energies are then used to rank order
2. METHOD AND RESULTS
2.1. SMARTCyp Score Functions for Ranking CYP3A4
and CYP2D6 SOM. SMARTCyp uses the following score
function to rank likely sites of CYP3A4 catalyzed reactions
Score_3A4 = E − 8*A(kJ/mol)
(1)
In this equation, E is an estimate of the activation energy
required for a CYP-catalyzed reaction at a specific atom of the
substrate. It is assigned to each atom of a substrate by matching
SMARTS patterns24 to a lookup table of energies derived from
density functional theory calculations. The accessibility
descriptor, A, is defined as the ratio between the longest
bond path from a given atom divided by the longest bond path
present in the whole molecule.
The SMARTCyp score function for ranking likely sites of
CYP2D6-mediated metabolism is
Score_2D6 = E + Span2End _correction + N +Dist _correcti
on
(2)
where N+Dist_correction = 6.7 × (8 - N+Dist) when N+Dist < 8,
and N+Dist_correction = 0 when N+Dist ≥ 8; Span2End_correction = 6.7 × Span2End when Span2End < 4, and
Span2End_correction = 6.7 × 4 + 0.01 × Span2End when
Span2End ≥ 4.
The Span2End descriptor accounts for the observation that
atoms in the middle of a compound are less likely to be
CYP2D6-mediated SOM than atoms at the ends of the
molecule. N+Dist is the maximum distance in number of
chemical bonds between a likely SOM to a protonated nitrogen
atom. It accounts for the fact that the Glu216 and/or Asp301
residues of CYP2D6 tend to form a characteristic salt bridge
with a protonated nitrogen atom in the substrate.25 Because the
Glu216 and Asp301 residues are located far away from the
catalytic heme moiety, substrate atoms close to the protonated
nitrogen atom are less likely to be CYP2D6-mediated SOM.
The constant, 6.7 kJ/mol, and the threshold values of N+Dist
and Span2End were determined using a training set of 86
CYP2D6 substrates.
2.2. SOM Prediction for CYP1A2-Catalyzed Reactions.
CYP1A2 constitutes approximately 14% of human liver P450
1699
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
enzymes in Caucasians26 and about 18% in Asians.4 It is
responsible for the clearance of about 5% of the top 200 drugs
on the U.S. market.27 Its catalytic site has been well
characterized.28−31 The X-ray structure of the protein
cocrystallized with α-naphthoflavone (ANF) indicates that the
substrate-binding cavity of CYP1A2 is narrow and lined by
amino acid residues that define a relatively planar substrate
binding platform.29 ANF is a potent and competitive inhibitor
of CYP1A2-catalyzed reactions. Its high binding affinity to
CYP1A2 was attributed to the overall fitness of the shape and
size of the molecule to the binding site cavity and the resulting
dense and extensive van der Waals interactions with the
nonpolar side chains of the protein. Additionally, it was noted
that there is a water molecule close to the carbonyl group of
ANF that provides an extra binding interaction. This water
molecule is the only one present in the active site, and there
appears to be no solvent channels that connect the active site
cavity with the protein surface. As shown in Figure 1, the water
molecule appears to be hydrogen-bonded to the carbonyl of
ANF as well as to the carbonyl of Gly316.29
derived from density functional theory calculations with
docking into the X-ray crystal structure of CYP1A2.32 For 60
CYP1A2 substrates, the accuracy of SOM prediction by the
reactivity + docking approach is similar to that of MetaSite
version 3.0, as shown in Table 1. Both the docking and
MetaSite calculations require 3D molecular structure information and evaluation of the fitness of substrate molecular shape
and size to the binding site cavity. Encouraged by the success of
SMARTCyp for CYP3A4 and CYP2D6 SOM predictions,
which use 2D molecular structure information of the substrates
only, we examined whether prediction accuracies similar to that
achieved by the reactivity/docking method and MetaSite could
be achieved for CYP1A2 by the SMARTCyp approach.
We first examined the performance of using SMARTCyp
reactivity only, without any substrate-CYP1A2 recognition
information, to predict SOM of CYP1A2 mediated metabolic
reactions. We examined two schemes of SMARTCyp reactivity
as defined by eq 1 and by the following equation
Score_2D6′ = E + Span2End _correction
(3)
Equation 1 is the scoring function for CYP3A4-mediated
SOM. Equation 3 is equal to the scoring function for CYP2D6catalyzed SOM minus the N+Dist_correction term. The
N+Dist_correction term was excluded in eq 3 because Rydberg
et al. used it to account for the effect of CYP2D6-substrate
recognition, which is not applicable in the case for CYP1A2.
Table 1 compares the performance of SOM predictions by
Score_3A4 and Score_2D6′, for the 60 CYP1A2 substrates
used by Rydberg et al. in their recent study,32 with the
performance of MetaSite and Rydberg’s reactivity + docking
approach. Molecular structures of the 60 CYP1A2 substrates,
observed SOM, and predicted top three SOM by Score_3A4
and Score_2D6′ are provided in Figure S1 in the Supporting
Information. Compared to the percentage of observed SOM in
randomly picked sites (column random in Table 1), both
Score_3A4 and Score_2D6′ performed reasonably well,
considering none of the information was CYP1A2-specific.
These results indicate that chemical reactivity is the most
important determinant for SOM of CYP1A2-catalyzed
metabolic reactions.
Table 1 also shows that Score_2D6′ outperformed
Score_3A4 in all four performance criteria (percentage of the
observed major sites of metabolism found in the top-ranked 1,
2, and 3 sites, as well as the percentage of observed major or
minor sites of metabolism found in the highest ranked sites). It
clearly indicates that Score_2D6′ is a better representation of
CYP1A2 substrate site reactivity than Score_3A4. This can be
rationalized by noting that the CYP3A4 substrate binding cavity
is large and very flexible, while the CYP1A2 binding site cavity
is much smaller and narrower, and, hence, the latter is more
similar to the substrate binding site cavity of CYP2D6.
Figure 1. The X-ray crystal structure of human CYP1A2 cocrystallized
with a potent inhibitor, α-naphthoflavone, showing tight binding due
to π−π stacking with residue Phe226 and a structured water molecule
in hydrogen bonding interactions with the ligand carbonyl group at
one end and with the carbonyl group of Gly316 at the other end. The
ligand-protein interactions orient the ligand so that the 4′ position of
the phenyl ring is exposed to the heme. For clarity, other parts of the
protein are not displayed.
Rydberg et al. showed that reliable SOM prediction for
CYP1A2 substrates can be achieved by combining site reactivity
Table 1. Performance Comparison for CYP1A2-Catalyzed Site of Metabolism (SOM) Prediction on 60 CYP1A2 Substratesa
Score_3A4b Score_2D6′c
number of compounds
1st ranked site is a major SOM
1st or 2nd ranked site is a major SOM
1st, 2nd, or 3rd ranked is a major SOM
1st ranked is a major or minor SOM
60
55
70
78
67
60
57
73
82
72
Score_2D6′
32 w/o rCO
72
84
91
81
28 w/rCO
39
61
71
61
Score_1A2d reactivity + dockinge MetaSitee
60
67
80
83
78
60
67
72
87
77
60
68
78
85
77
random
60
15
29
42
19
a
c
Numerical values in the table are percentages of the substrates for which the SOM were correctly ranked. bScore_3A4 is defined by eq 1.
Score_2D6′ is defined by eq 3. dScore_1A2 is defined by eq 5. eReference 32.
1700
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 2. continued
1701
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 2. SOM of CYP1A2-catalyzed reactions. Red arrows: observed major sites. Blue arrows: observed minor sites. Red numbers: top-ranked sites
by Score_1A2. For molecules with symmetrically equivalent sites, only one of the equivalent sites is labeled.
However, even with Score_2D6′, the performance of SOM
prediction for CYP1A2 substrates was generally inferior to that
of MetaSite or the reactivity + docking approach. Detailed
examination of the structures of the 60 substrates revealed that
1702
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 3. (A) X-ray structure of human CYP2C9 cocrystallized with flurbiprofen showing hydrogen bonding interactions between the anionic
carboxyl group with the Asn204 and Arg108 side chains of the protein. For clarity, other parts of the protein are not displayed. (B) Molecular
structure of n-undecane optimized by the MMFF force field.
28 of them have a carbonyl group with the carbon atom being
part of a ring. For the 32 molecules without a cyclic carbonyl
moiety, SOM prediction given by Score_2D6′ significantly
outperformed that of MetaSite and the reactivity + docking
approach, as shown in Table 1. However, for the 28
compounds with a cyclic carbonyl moiety, SOM prediction
given by Score_2D6′ alone was significantly worse. This
indicates that the cyclic carbonyl group may be an important
contributor to the SOM of CYP1A2-catalyzed reactions. It
could be a key substrate recognition feature of the CYP1A2
enzyme. The substrate-enzyme recognition feature may orient
the substrate so that some reactive sites of the substrate are out
of reach by the catalytic heme moiety. Figure 1 shows a likely
interaction between the carbonyl group of a substrate and the
enzyme - the hydrogen bonding interaction between the
carbonyl oxygen and a water molecule. The water molecule, in
turn, forms hydrogen bonding interactions with the carbonyl
oxygen of Gly316 of the protein, as reported by Sansen et al.29
This is an example of a water molecule serving as a bridge for
receptor−ligand recognition. The fact that a ring carbonyl is
required for this receptor−ligand interaction can be explained
by the narrow and flat binding cavity of the enzyme. A
noncyclic carbonyl group may not be as effectively anchored to
the binding site because of steric hindrance from the left and
right groups attached to the carbonyl carbon atom. The ring
ties the two groups back, making the carbonyl oxygen
effectively exposed for hydrogen bonding interactions with
the water molecule. The ring moiety fits the flat binding cavity
well and forms favorable van der Waals interactions with the
hydrophobic side chains of the protein.
Figure 1 shows that, because of the ring carbonyl-water
hydrogen bonding interactions, carbon atoms at the 3′, 4′, and
5′ positions of the phenyl ring are about 5 Å from the heme
iron, a distance generally considered to be within the reach of
the FeO3+ catalytic moiety of the heme.33 On the other hand,
atoms within four chemical bonds of the carbonyl carbon atom
appear too far from the catalytic moiety. However, this is only a
static picture of a protein-bound ligand. In reality, both the
protein and the ligand have some degrees of freedom. In light
of the SMARTCyp approach for predicting SOM of CYP2D6mediated metabolic reactions, we introduced a correction term
to account for the effect of substrate-CYP1A2 recognition. This
term is called Dist2rCO_correction and is defined as
Dist2rCO _correction = c × (Dist2rCO _cutoff − Dist2rCO)
(4)
In eq 4, Dist2rCO denotes the distance in number of
chemical bonds between a substrate atom of interest and the
most distant cyclic carbonyl carbon of the substrate. The
Dist2rCO_cutof f is a threshold bond distance above which no
correction is needed. Equation 4 is similar to the
N+Dist_correction term in eq 2 for CYP2D6-catalyzed SOM.
Rydberg and Olsen selected the constant c to be 6.7 kJ/mol and
N+Dist_cutof f to be 8 for CYP2D6 by using a training set of 86
CYP2D6 substrates. In the case of CYP1A2, we do not have a
large number of substrates as a training set for determining
these constants. However, we found that the results were not
very sensitive to the value of the constant c, as long as it was
approximately 10 kJ/mol. Therefore, for simplicity, we assigned
c = 10 kJ/mol and experimented with bond distance cutoff
values of Dist2rCO_cutof f. For the 28 substrates with cyclic
carbonyl groups, the best performance was achieved with
Dist2rCO_cutof f = 6. In the end, our score function for ranking
likely SOM for CYP1A2-catalyzed metabolic reactions was
Score_1A2 = E + Span2End _correction + Dist2rCO _correc
tion
(5)
where Dist2rCO_correction = 10.0 × (6 − Dist2rCO) when
Dist2rCO < 6, and Dist2rCO_correction = 0 when Dist2rCO ≥
6.
With the Dist2rCO_correction term to account for the effect
of CYP1A2 - substrate recognition, the SOM prediction based
on Score_1A2 improved significantly. Performance of SOM
prediction for the 60 substrates is shown in Table 1 under the
column Score_1A2. The molecular structures of the 60
substrates, observed SOM, and predicted top three SOM by
Score_1A2 are given in Figure 2.
Based on all four criteria in Table 1, the 2D ligand-based
Score_1A2 is at least as good as the 3D structure-based
MetaSite and the reactivity + docking approach, indicating that
the water mediated substrate recognition feature is an
important determinant for CYP1A2-catalyzed metabolic
reactions.
2.3. SOM Prediction for CYP2C9-Catalyzed Reactions.
According to Rowland-Yeo et al., nearly 20% of the CYP
enzymes in the human liver are CYP2C9.26 CYP2C9
1703
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
metabolizes about 15% of marketed drugs27 and is known to
exhibit selectivity for the oxidation of relatively small, lipophilic
anions.34 Figure 3 shows the X-ray crystal structure of human
CYP2C9 cocrystallized with flurbiprofen. The structure
indicates that the carboxyl group of flurbiprofen forms
hydrogen bonding interactions with the Arg108 and Asn204
side chains of the protein.35 Since the Arg108 and Asn204 side
chains are at the opposite end to the heme in the substrate
binding cavity, substrate atoms in close proximity to the
carboxyl group that forms hydrogen bonding interactions with
the side chains are less likely to be the SOM because of their
distance from the catalytic moiety. The high resolution crystal
structure showed that the carboxyl carbon atom of flurbiprofen
is 13.2 Å from the heme iron. The distance is close to 10 C−C
single bonds (∼12.6 Å), as shown in Figure 3. It is generally
considered that substrate atoms within ∼5 Å of the heme iron
are accessible by the FeO3+ catalytic moiety and may become
the site of CYP-mediated metabolic reactions.33 Based on the
MMFF force field calculations, 5 Å is approximately the
distance between the terminal carbon atoms in n-pentane, or
four C−C bonds, as shown in Figure 3(B). Taken together,
Figure 3 indicates that substrate atoms within five chemical
bonds of the carboxyl group have reduced probability to be
sites of CYP2C9 catalyzed metabolic reactions. Furthermore,
the closer an atom is to the carboxyl group, the less likely it is to
be metabolized by CYP2C9. On the basis of this key substratereceptor recognition feature and consistent with the SMARTCyp approach, a reasonable score function for ranking likely
sites of CYP2C9 metabolism is
score function as measured by the percentage of cases of the
experimentally observed SOM were among the top-ranked 1, 2,
and 3 sites. For comparison, the performance of SOM
prediction by StarDrop version 5.0 is also included in Table
2. By all three performance measures given in Table 2, the score
function of eq 6 outperformed StarDrop 5.0 for the 21
carboxylic acid substrates.
To evaluate if it is necessary to apply the progressive energy
penalty for atoms in close proximity to the carboxyl group, we
also made SOM predictions using Score_2D6′ only. The
corresponding values of the correctly predicted SOM for the 21
carboxylic acids are 52%, 67%, and 67%, respectively. The
significantly inferior results obtained by using Score_2D6′
underscores the importance of the key substrate - receptor
recognition in determining the regioselectivity of CYP2C9catalyzed metabolic reactions. These results also validate the
use of Dist2CO_correction to account for effects of this substrate
- receptor recognition.
We also applied Score_2D6′ to rank likely sites of CYP2C9
metabolism for the 49 noncarboxylic substrates collated by
Sykes et al. For these compounds, the percentages of the
observed SOM being among the top-ranked 1, 2, and 3 sites
were 57%, 76%, and 80%, respectively. The results are close to
the performance of StarDrop 5.0 for these compounds but
significantly worse than the performance of Score_2C9 for the
carboxylic acids. Inspection of the molecular structures of the
compounds revealed that, even though they are noncarboxylic
acids, many of them have carbonyl groups. Carbonyl groups are
perhaps weaker hydrogen bond acceptors than the negatively
charged carboxyl groups, but nonetheless, they are hydrogen
bond acceptors. There is no reason to believe that they do not
form hydrogen bonding interactions with the Arg108 or
Asn204 side chains of the CYP2C9 enzyme. Failing to account
for this hydrogen bonding interaction might be responsible for
the inferior performance of the score function for noncarboxylic
substrates. To test this hypothesis, we redefined Dist2CO in eq
6. For a carboxylic acid or anion, it does not matter if there are
other carbonyl groups or not, Dist2CO is always defined as the
distance in number of chemical bonds between a substrate
atom of interest and the most distant carboxylic carbon atom of
the substrate. For other substrates, Dist2CO is defined as the
distance in number of chemical bonds between a substrate
atom of interest and the most distant carbonyl carbon of the
substrate. Note that for a compound with both carboxyl and
carbonyl groups, the carboxyl group is given precedence in
calculating Dist2CO. This is based on the observation that the
carboxyl group in Figure 3 forms hydrogen bonding
interactions with both the Arg108 and Asn204 residues of the
protein. A noncarboxyl carbonyl group most likely forms
weaker hydrogen bonding interactions with only one of the two
protein residues. This is consistent with the observation that
CYP2C9 exhibits selectivity for the oxidation of carboxylic acid
substrates. With this modification, Score_2C9 was applied to
rank likely SOM of the 49 noncarboxylic acid substrates. Table
2 shows that the results were significantly improved. Overall,
for the 70 CYP2C9 substrates, the observed SOM were found
in the top-ranked 1, 2, and 3 sites in 66%, 86%, and 87% of the
cases, respectively. For comparison, StarDrop 5.0 was also used
to predict SOM for the 70 compounds. However, our StarDrop
calculations were successful for only 69 of the 70 compounds.
The calculation for zafirlukast failed, presumably due to some
problem in the semiempirical AM1 stage. The corresponding
percentages for the 69 compounds achieved by StarDrop were
Score_2C9 = E + Span2End _correction + Dist2CO _correct
(6)
ion
where Dist2CO_correction =10 × (6 − Dist2CO) when Dist2CO
≤ 5, and Dist2CO_correction = 0 when Dist2CO ≥ 6.
In eq 6, Dist2CO is the distance in number of chemical bonds
between a substrate atom of interest and the most distant
carboxyl carbon of the substrate. The Dist2CO_correction term
accounts for the effect of substrate - receptor recognition on
CYP2C9 SOM.
To test this scoring function, we applied it to predict the
SOM of 21 carboxylic acids collated by Sykes et al.36 that are
metabolized by CYP2C9. Table 2 gives the performance of the
Table 2. Performance Comparison for CYP2C9-Catalyzed
Site of Metabolism (SOM) Predictions on 70 CYP2C9
Substratesa
Score_2C9b
1st ranked site is
the observed
SOM
1st or 2nd ranked
site is the
observed SOM
1st, 2nd, or 3rd
ranked site is
the observed
SOM
StarDrop 5.0c
carboxylic
acids
nonacid
overall
carboxylic
acids
nonacid
overall
67
65
66
57
63
61
81
88
86
71
73
73
86
88
87
71
79
77
a
Numerical values in the table are percentages of the substrates for
which the observed SOM were correctly ranked. bScore_2C9 is
defined eq 6. cPrediction given by StarDrop version 5.0 on 69 of the
70 substrates. StarDrop calculation on one of the 70 compounds failed.
1704
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 4. continued
1705
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 4. continued
1706
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 4. SOM of CYP2C9-catalyzed reactions. Red arrows: observed sites of metabolism. Red numbers: top-ranked sites by Score_2C9. For
molecules with symmetrically equivalent sites, only one of the equivalent sites is labeled.
61%, 73%, and 77%, respectively. Figure 4 shows molecular
structures of the 70 CYP2C9 substrates, the experimentally
observed SOM, and the three top-ranked sites by Score_2C9.
The fact that the redefined Dist2CO_correction term improved
SOM prediction supports the hypothesis that carbonyl groups
also form hydrogen bonding interactions with CYP2C9, which
influences the SOM of CYP2C9-catalyzed metabolic reactions.
2.4. SOM Prediction for CYP2C19-Catalyzed Metabolic Reactions. CYP2C9 and CYP2C19 have a very high level of
sequence identity. The two proteins have only 43 differing
amino acid residues among a total sequence length of 490
residues.37 However, they have quite different substrate
specificity. For instance, while CYP2C9 shows selectivity for
and is a major metabolizer of acidic substrates that are anionic
under physiological pH, most signature CYP2C19 substrates
are lipophilic and neutral at physiological pH.38 In addition,
CYP2C19 is highly selective for 4′-hydroxlation of mephenytoin
and 5-hydroxylation of proton pump inhibitors such as
omeprazole and lansoprazole, while CYP2C9 has little activity
for these compounds.39 Because of this, mephenytoin,
omeprazole, and lansoprazole were termed marker substrates
of CYP2C19 by Wada et al.39
Even though a crystal structure of human CYP2C19 is not
yet available, mutation studies have identified some key amino
acids that are crucial for conferring CYP2C9 and CYP2C19
substrate specificity.40 For example, molecular structural
analysis indicated that the F-G loop forms a flexible lid and a
substrate entrance channel in CYP2C8, CYP2C9, and
CYP2B4.41−43 Substitution of the F-G loop in CYP2C9 to
that of CYP2C19, i.e., Ser220 → Pro and Pro221 → Thr, does
not alter the enzyme activity toward CYP2C9 marker substrates
but enhanced 4′-hydroxylation of mephenytoin and 5hydroxylation of omeprazole, which were not detectable in
CYP2C9.39 In addition, mutation of Ile99 to His99 (of
CYP2C19) in CYP2C9 increased omeprazole 5-hydroxylase
to ∼51% of that of CYP2C19.40 This provides strong evidence
that amino acids 99, 220, and 221 are among the key residues
that determine the distinctive substrate specificities of CYP2C9
and CYP2C19.
The crystal structure of CYP2C9 indicates that Ile99 is very
close to the heme,35 and it is expected that this is also the case
for His99 in CYP2C19. Locuson et al. postulated that His99
plays an important role in CYP2C19 metabolism by serving as a
hydrogen bond donor and forming hydrogen bonding
1707
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 5. Hydrogen bonding interactions, postulated by Locuson et al., between His99 of CYP2C19 and a substrate, and the role of the hydrogen
bonding interactions in the regioselectivity of the metabolic reactions. The sketch was adapted with permission from the paper of Locuson et al.
published in J. Med. Chem. 2004, 47, 6768−6776 Copyright 2004, American Chemical Society.
Test calculations indicated that a Cutof f value of 4 is
reasonable. This threshold value implies that substrate atoms
within three chemical bonds of CO or SO are likely to be
positioned too far from the catalytic heme moiety and,
therefore, have lower probabilities to be sites of CYP2C19catalyzed reactions. Figure 6 shows the top three sites for each
substrate ranked by Score_2D6′ and Score_2C19. Overall, an
observed SOM was found to be among the top 1, 2, and 3 sites
ranked by Score_2C19 78%, 89%, and 94% of the time,
respectively, for the 36 substrates. Two factors contributed to
the higher percentage of correct SOM predictions for
CYP2C19 substrates than for CYP1A2 and CYP2C9 substrates.
First, the number of CYP2C19 substrates we collected and used
in the study was relatively small. Second, and more importantly,
the observed metabolic reactions for approximately half of the
CYP2C19 substrates were either N- or O-dealkylation
reactions. These reactions have significantly lower activation
barriers than most other reactions. As a result, the SOM of Nand O-dealkylation reactions were correctly predicted by
SMARTCyp energies for most of the compounds.
interactions with the carbonyl or sulfinyl oxygen of the
substrate, as illustrated in Figure 5.44 The hydrogen bonding
interactions enhance the affinity of the substrates to the enzyme
and orient the substrates to the nearby heme catalytic site.
To examine if the SMARTCyp approach could be extended
to predict SOM of CYP2C19 mediated metabolic reactions, we
collated 36 compounds metabolized by CYP2C19. The
observed CYP2C19 SOM of these compounds were obtained
from the papers of Lewis et al.38 and Locuson et al.44 and from
DrugBank.45
As a first step, we applied Score_2D6′ to evaluate if
SMARTCyp reactivity alone was sufficient for predicting
SOM of the 36 CYP2C19 substrates. As shown in Figure 6
and Table 3, reasonably reliable predictions were achieved by
Score_2D6′ without using any CYP2C19-specific information.
The observed sites of CYP2C19 metabolism were found to be
among the top-ranked 1, 2, and 3 sites in 69%, 89%, and 92% of
the cases, respectively. However, for the CYP2C19 marker
substrates mephenytoin and lansoprazole, the observed sites of
metabolism were second-ranked sites by Score_2D6′, as shown
in Figure 6. To improve prediction consistent with the
SMARTCyp approach, we adopted the following score function
3. CONCLUSIONS
The results of this study, together with those of Rydberg et al.
on CYP3A4 and CYP2D6 SOM predictions, demonstrated that
chemical reactivity is the most important determinant of
substrate SOM of CYP-catalyzed reactions.They also demonstrated that the activation energies of Rydberg et al., derived
from density functional theory calculations and corrected by
the Span2End descriptor, are reasonable representations of
substrate site reactivity for most CYP enzymes. Highly reliable
SOM predictions for CYP1A2, CYP2C9, and CYP2C19 were
derived based on the reactivity combined with a correction
term to account for the effect of key substrate - enzyme
recognition on substrate access to the catalytic moiety. The fact
Score_2C19 = E + Span2End _correction + Dist2XO _corre
ction
(7)
Dist2XO_correction = 10 × (Cutof f − Dist2XO), when Dist2XO
< Cutof f, and Dist2XO_correction = 0 otherwise.
In the preceding equations, XO represents CO or SO
moieties, as they were postulated by Locuson et al.44 to be
CYP2C19 substrate recognition features. Dist2XO denotes the
distance in number of chemical bonds between a substrate
atom of interest to the most distant CO or SO group of
the substrate.
1708
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 6. continued
1709
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Figure 6. SOM of CYP2C19-catalyzed reactions. Red arrows: observed sites of metabolic reactions. Blue numbers: top-ranked sites by Score_2D6′
(no CYP2C19 specific information is used). Red numbers: top-ranked sites by Score_2C19. For molecules with symmetrically equivalent sites, only
one of the equivalent sites is labeled.
anyone without the need for specialized training. The score
functions for ranking likely sites of metabolism by CYP1A2,
2C9, and 2C19 enzymes as described in this article were
implemented as an extension of the SMARTCyp 2.0 program
by modifying the program modules. With the extension to
cover the three additional CYP isoforms, the program can now
successfully predict SOM for all five major drug metabolizing
CYP enzymes. Both the Java source code and the binary
executable of the program are freely available for download at
http://www.bhsai.org/downloads/smartcyp_ext/.
Table 3. Performance Comparison for CYP2C19-Catalyzed
Site of Metabolism (SOM) Predictions on 36 CYP2C19
Substratesa
1st ranked site is the observed SOM
1st or 2nd ranked site is the observed SOM
1st, 2nd, or 3rd ranked site is the observed
SOM
Score_2D6′b
Score_2C19c
69
89
92
78
89
94
a
Numerical values in the table are percentages of the substrates for
which the observed SOM were correctly ranked. bScore_2D6′ is
defined by eq 3. cScore_2C19 is defined by eq 7.
■
ASSOCIATED CONTENT
S Supporting Information
*
that the correction term defined by the distance to a cyclic
carbonyl group significantly improved CYP1A2 SOM predictions supports the finding that a structured water molecule
in the CYP1A2 substrate binding cavity plays an important role
in CYP1A2 substrate recognition. This water molecule was
observed to be hydrogen-bonded to the carbonyl of a CYP1A2
inhibitor and to the carbonyl of Gly316 of the enzyme.
Similarly, the fact that a correction term, defined by the
distance to a substrate carbonyl or sulfinyl group, improved
CYP2C19 SOM prediction supports the hypothesis of Locuson
et al. on the role of residue His99 in CYP2C19 metabolism.
In principle, docking of substrates into the protein structures
should give reliable prediction of substrate sites that may be
accessible to the catalytic moiety. However, docking
simulations are much more computationally demanding than
the 2D substrate structure based SMARTCyp approach for
SOM prediction. In addition, it is challenging to properly
account for protein flexibility in a docking simulation, and it
requires expert knowledge to select an appropriate docking
score function to rank the docked poses. As a result, docking is
more of an expert tool that requires significant knowledge and
experience to process computational results and select relevant
poses from a large number of docked conformations. On the
other hand, the SMARTCyp approach for predicting likely sites
of CYP-catalyzed metabolic reactions is very fast as it uses 2D
molecular structure information only and is easily applied by
Figure S1. SOM of CYP1A2 catalyzed reactions predicted by
Score_3A4 (eq 1) and by Score_2D6′ (eq 3). This material is
available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*Phone: 301-619-1979. Fax: 301-619-1983. E-mail: rliu@bhsai.
org.
Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS
Funding of this research was provided by the U.S. Department
of Defense Threat Reduction Agency Grant
TMTI0004_09_BH_T. The opinions or assertions contained
herein are the private views of the authors and are not to be
construed as official or as reflecting the views of the U.S. Army
or of the U.S. Department of Defense. We thank Mr. Li Chen
and Jason Smith for their technical assistance. This paper has
been approved for public release with unlimited distribution.
■
REFERENCES
(1) Sigel, A.; Sigel, H.; Sigel, R. The Ubiquitous Roles of Cytochrome
P450 Proteins: Metal Ions in Life Sciences; Wiley: 2007; p 3.
1710
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
Substrates Using MetaSite and StarDrop. Comb. Chem. High
Throughput Screening 2011, 14, 811−823.
(23) Rydberg, P.; Olsen, L. Ligand-Based Site of Metabolism
Prediction for Cytochrome P450 2D6. ACS Med. Chem. Lett. 2012, 3,
69−73.
(24) Daylight Chemical Information Systems, I., SMARTS - A
language for describing molecular patterns, http://www.daylight.com/
dayhtml/doc/theory/theory.smarts.html (accessed May 31, 2012).
(25) Guengerich, F. P.; Hanna, I. H.; Martin, M. V.; Gillam, E. M.
Role of glutamic acid 216 in cytochrome P450 2D6 substrate binding
and catalysis. Biochemistry 2003, 42, 1245−1253.
(26) Rowland-Yeo, K.; Rostami-Hodjegan, A.; Tucker, G. T.
Abundance of cytochromes P450 in human liver: a meta-analysis. Br.
J. Clin. Pharmacol. 2004, 57, 687−688.
(27) Williams, J. A.; Hyland, R.; Jones, B. C.; Smith, D. A.; Hurst, S.;
Goosen, T. C.; Peterkin, V.; Koup, J. R.; Ball, S. E. Drug-drug
interactions for UDP-glucuronosyltransferase substrates: a pharmacokinetic explanation for typically observed low exposure (AUCi/AUC)
ratios. Drug Metab. Dispos. 2004, 32, 1201−1208.
(28) Ekins, S.; de Groot, M. J.; Jones, J. P. Pharmacophore and threedimensional quantitative structure activity relationship methods for
modeling cytochrome p450 active sites. Drug Metab. Dispos. 2001, 29,
936−944.
(29) Sansen, S.; Yano, J. K.; Reynald, R. L.; Schoch, G. A.; Griffin, K.
J.; Stout, C. D.; Johnson, E. F. Adaptations for the oxidation of
polycyclic aromatic hydrocarbons exhibited by the structure of human
P450 1A2. J. Biol. Chem. 2007, 282, 14348−14355.
(30) Smith, D. A.; Ackland, M. J.; Jones, B. C. Properties of
cytochrome P450 isoenzymes and their substrates part 1: active site
characteristics. Drug Discovery Today 1997, 2, 406−478.
(31) Smith, D. A.; Ackland, M. J.; Jones, B. C. Properties of
cytochrome P450 isoenzymes and their substrates part 2: properties of
cytochrome P450 substrates. Drug Discovery Today 1997, 2, 479−486.
(32) Rydberg, P.; Vasanthanathan, P.; Oostenbrink, C.; Olsen, L. Fast
prediction of cytochrome P450 mediated drug metabolism. ChemMedChem 2009, 4, 2070−2079.
(33) Prasad, J. C.; Goldstone, J. V.; Camacho, C. J.; Vajda, S.;
Stegeman, J. J. Ensemble modeling of substrate binding to
cytochromes P450: analysis of catalytic differences between CYP1A
orthologs. Biochemistry 2007, 46, 2640−2654.
(34) Tai, G.; Dickmann, L. J.; Matovic, N.; DeVoss, J. J.; Gillam, E.
M.; Rettie, A. E. Re-engineering of CYP2C9 to probe acid-base
substrate selectivity. Drug Metab. Dispos. 2008, 36, 1992−1997.
(35) Wester, M. R.; Yano, J. K.; Schoch, G. A.; Yang, C.; Griffin, K. J.;
Stout, C. D.; Johnson, E. F. The structure of human cytochrome P450
2C9 complexed with flurbiprofen at 2.0-A resolution. J. Biol. Chem.
2004, 279, 35630−35637.
(36) Sykes, M. J.; McKinnon, R. A.; Miners, J. O. Prediction of
metabolism by cytochrome P450 2C9: alignment and docking studies
of a validated database of substrates. J. Med. Chem. 2008, 51, 780−791.
(37) Jung, F.; Griffin, K. J.; Song, W.; Richardson, T. H.; Yang, M.;
Johnson, E. F. Identification of amino acid substitutions that confer a
high affinity for sulfaphenazole binding and a high catalytic efficiency
for warfarin metabolism to P450 2C19. Biochemistry 1998, 37, 16270−
16279.
(38) Lewis, D. F.; Dickins, M.; Weaver, R. J.; Eddershaw, P. J.;
Goldfarb, P. S.; Tarbit, M. H. Molecular modelling of human CYP2C
subfamily enzymes CYP2C9 and CYP2C19: rationalization of
substrate specificity and site-directed mutagenesis experiments in the
CYP2C subfamily. Xenobiotica 1998, 28, 235−268.
(39) Wada, Y.; Mitsuda, M.; Ishihara, Y.; Watanabe, M.; Iwasaki, M.;
Asahi, S. Important amino acid residues that confer CYP2C19 selective
activity to CYP2C9. J. Biochem. 2008, 144, 323−333.
(40) Ibeanu, G. C.; Ghanayem, B. I.; Linko, P.; Li, L.; Pederson, L.
G.; Goldstein, J. A. Identification of residues 99, 220, and 221 of
human cytochrome P450 2C19 as key determinants of omeprazole
activity. J. Biol. Chem. 1996, 271, 12496−12501.
(41) Schoch, G. A.; Yano, J. K.; Wester, M. R.; Griffin, K. J.; Stout, C.
D.; Johnson, E. F. Structure of human microsomal cytochrome P450
(2) Danielson, P. B. The cytochrome P450 superfamily: biochemistry, evolution and drug metabolism in humans. Curr. Drug. Metab.
2002, 3, 561−597.
(3) Guengerich, F. P. Cytochrome p450 and chemical toxicology.
Chem. Res. Toxicol. 2008, 21, 70−83.
(4) Shimada, T.; Yamazaki, H.; Mimura, M.; Inui, Y.; Guengerich, F.
P. Interindividual variations in human liver cytochrome P-450 enzymes
involved in the oxidation of drugs, carcinogens and toxic chemicals:
studies with liver microsomes of 30 Japanese and 30 Caucasians. J.
Pharmacol. Exp. Ther. 1994, 270, 414−423.
(5) Lynch, T.; Price, A. The effect of cytochrome P450 metabolism
on drug response, interactions, and adverse effects. Am. Fam. Physician
2007, 76, 391−396.
(6) Slaughter, R. L.; Edwards, D. J. Recent advances: the cytochrome
P450 enzymes. Ann. Pharmacother. 1995, 29, 619−624.
(7) Wilkinson, G. R. Drug metabolism and variability among patients
in drug response. New Engl. J. Med. 2005, 352, 2211−2221.
(8) Guengerich, F. P. Common and uncommon cytochrome P450
reactions related to metabolism and chemical toxicity. Chem. Res.
Toxicol. 2001, 14, 611−650.
(9) de Graaf, C.; Oostenbrink, C.; Keizers, P. H.; van der Wijst, T.;
Jongejan, A.; Vermeulen, N. P. Catalytic site prediction and virtual
screening of cytochrome P450 2D6 substrates by consideration of
water and rescoring in automated docking. J. Med. Chem. 2006, 49,
2417−2430.
(10) de Groot, M. J.; Ackland, M. J.; Horne, V. A.; Alex, A. A.; Jones,
B. C. Novel approach to predicting P450-mediated drug metabolism:
development of a combined protein and pharmacophore model for
CYP2D6. J. Med. Chem. 1999, 42, 1515−1524.
(11) Park, J. Y.; Harris, D. Construction and assessment of models of
CYP2E1: predictions of metabolism from docking, molecular
dynamics, and density functional theoretical calculations. J. Med.
Chem. 2003, 46, 1645−1660.
(12) Vasanthanathan, P.; Hritz, J.; Taboureau, O.; Olsen, L.;
Jorgensen, F. S.; Vermeulen, N. P.; Oostenbrink, C. Virtual screening
and prediction of site of metabolism for cytochrome P450 1A2 ligands.
J. Chem. Inf. Model. 2009, 49, 43−52.
(13) Yuriev, E.; Agostino, M.; Ramsland, P. A. Challenges and
advances in computational docking: 2009 in review. J. Mol. Recognit.
2011, 24, 149−164.
(14) Moors, S. L.; Vos, A. M.; Cummings, M. D.; Van Vlijmen, H.;
Ceulemans, A. Structure-based site of metabolism prediction for
cytochrome P450 2D6. J. Med. Chem. 2011, 54, 6098−6105.
(15) Afzelius, L.; Arnby, C. H.; Broo, A.; Carlsson, L.; Isaksson, C.;
Jurva, U.; Kjellander, B.; Kolmodin, K.; Nilsson, K.; Raubacher, F.;
Weidolf, L. State-of-the-art tools for computational site of metabolism
predictions: comparative analysis, mechanistical insights, and future
applications. Drug Metab. Rev. 2007, 39, 61−86.
(16) Singh, S. B.; Shen, L. Q.; Walker, M. J.; Sheridan, R. P. A model
for predicting likely sites of CYP3A4-mediated metabolism on druglike molecules. J. Med. Chem. 2003, 46, 1330−1336.
(17) StarDrop, version 5.0; Optibrium Ltd.: Cambridge, United
Kingdom, 2011.
(18) Cruciani, G.; Carosati, E.; De Boeck, B.; Ethirajulu, K.; Mackie,
C.; Howe, T.; Vianello, R. MetaSite: understanding metabolism in
human cytochromes from the perspective of the chemist. J. Med. Chem.
2005, 48, 6970−6979.
(19) Goodford, P. J. A computational procedure for determining
energetically favorable binding sites on biologically important
macromolecules. J. Med. Chem. 1985, 28, 849−857.
(20) Zhou, D.; Afzelius, L.; Grimm, S. W.; Andersson, T. B.; Zauhar,
R. J.; Zamora, I. Comparison of methods for the prediction of the
metabolic sites for CYP3A4-mediated metabolic reactions. Drug
Metab. Dispos. 2006, 34, 976−983.
(21) Rydberg, P.; Gloriam, D. E.; Zaretzki, J.; Breneman, C.; Olsen,
L. SMARTCyp: A 2D Method for Prediction of Cytochrome P450Mediated Drug Metabolism. ACS Med. Chem. Lett. 2010, 1, 96−100.
(22) Shin, Y. G.; Le, H.; Khojasteh, C.; Hop, C. E. Comparison of
Metabolic Soft Spot Predictions of CYP3A4, CYP2C9 and CYP2D6
1711
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712
Journal of Chemical Information and Modeling
Article
2C8. Evidence for a peripheral fatty acid binding site. J. Biol. Chem.
2004, 279, 9497−9503.
(42) Scott, E. E.; He, Y. A.; Wester, M. R.; White, M. A.; Chin, C. C.;
Halpert, J. R.; Johnson, E. F.; Stout, C. D. An open conformation of
mammalian cytochrome P450 2B4 at 1.6-A resolution. Proc. Natl.
Acad. Sci. U.S.A. 2003, 100, 13196−13201.
(43) Williams, P. A.; Cosme, J.; Sridhar, V.; Johnson, E. F.; McRee,
D. E. Mammalian microsomal cytochrome P450 monooxygenase:
structural adaptations for membrane binding and functional diversity.
Mol. Cell 2000, 5, 121−131.
(44) Locuson, C. W., 2nd; Suzuki, H.; Rettie, A. E.; Jones, J. P.
Charge and substituent effects on affinity and metabolism of
benzbromarone-based CYP2C19 inhibitors. J. Med. Chem. 2004, 47,
6768−6776.
(45) Knox, C.; Law, V.; Jewison, T.; Liu, P.; Ly, S.; Frolkis, A.; Pon,
A.; Banco, K.; Mak, C.; Neveu, V.; Djoumbou, Y.; Eisner, R.; Guo, A.
C.; Wishart, D. S. DrugBank 3.0: a comprehensive resource for ’omics’
research on drugs. Nucleic Acids Res. 2011, 39 (Database issue),
D1035−41.
1712
dx.doi.org/10.1021/ci3001524 | J. Chem. Inf. Model. 2012, 52, 1698−1712